AI Guides › Playbooks
By Nigel Guy · 7 min read
Most AI tool rankings are someone else's week. A creator lists the ten tools they "can't live without", you recognise four, sign up for two more, and a month later you are paying for three overlapping chat assistants and opening one of them. The list felt useful because it was confident. It was never about your work.
The rule: rank a tool by the hours of your real week it earns, not by its reputation, and cancel anything that cannot name the job it does better than the tool above it.
This playbook gives you the mechanism for doing that: the Earned-Week Ranking, a one-page table you fill from evidence (your bank statement and your last fortnight of work), not from memory or headlines.
For most people the honest answer is: in one general assistant, plus a small number of specialists. The general assistant is wherever you draft, think, summarise and ask "what am I missing?" Everything else should have to justify itself against that tool.
That matters for money. At time of writing, the flagship individual tiers of the big general assistants cluster at the same price point: Anthropic lists Claude Pro at $20 a month billed monthly (US pricing, charged in local currency where supported), and OpenAI lists ChatGPT Plus at $20 a month. UK checkout prices are shown in pounds and can include VAT, so check your own statement rather than converting. Paying for two or three of these is the most common overlap, and the easiest saving.
A specialist earns a slot only if it does one job the general assistant does badly or cannot do at all. The categories that tend to pass that test:
| Specialist category | The job it should own | The catch |
|---|---|---|
| Cited research search | Finding and linking current sources fast | Citations still need opening; a link is not a check |
| Meeting transcription | Turning calls into notes and actions | Consent and data-handling for everyone on the call |
| Image or video generation | Visual drafts you would otherwise commission | Credit-based pricing that is easy to overspend |
| Coding agent | Changes across real files, run and tested | Needs supervision; can burn usage quickly |
| Automation or connector tool | Moving data between apps without you | Silent failures; someone has to check the logs |
Not on the list: "a second chatbot that might write better". If it owns no job, it is a subscription, not a specialist.
Open your bank or card statement for the last two months and list every AI-related charge, including annual plans and app-store subscriptions. Then add the free tools you used at least once in the last fortnight. Memory flatters tools you like and forgets ones that bill quietly.
Use this table. Every column is a fact you can check, except the last, which is a decision.
| Tool | Cost per month (£, from statement) | The job it did, named | Uses in last 14 days | Would you notice if it vanished tomorrow? | Overlaps with | Verdict |
|---|---|---|---|---|---|---|
| Yes / Mildly / No | Keep / Downgrade / Cancel / Trial ends |
"The job it did" must be a verb and an object: "drafted client proposals", "transcribed Tuesday stand-ups". "Brainstorming" and "general stuff" do not count; they are the answer every tool gives.
Sort the rows by how much of a normal working week each tool genuinely carries. Your top slot is almost always your general assistant. Below it, each tool must pass the overlap test: is something higher on the list already doing this job well enough? If yes, it drops to Downgrade or Cancel, however much you enjoy it.
For each Cancel or Downgrade, write the date you will do it. For anything you are keeping on probation, write the job it must do in the next 30 days to stay. Then put a reminder in your calendar to rerun the table in three months, because plans, limits and your own work all change.
Once your table is filled in, paste it into your general assistant with the prompt below. Fill in: your tool list with costs, what you actually used each for, and your typical working week.
You are a sceptical operations adviser helping one person cut their AI tool spending. You have no loyalty to any vendor and you do not recommend new tools unless I ask.
My context:
- My work, in one or two sentences: [YOUR_ROLE_AND_TYPICAL_WEEK]
- My tools, each with monthly cost in pounds: [TOOL_LIST_WITH_COSTS]
- What I actually used each one for in the last two weeks, in my words: [ACTUAL_USES]
Goal: an honest ranking of these tools by how much of my real week each one carries, plus a clear verdict on each.
Steps:
1. If any tool is missing a cost or a described use, stop and ask me for it before ranking. Do not fill gaps with what the tool is "known for".
2. For each tool, name the single job it is doing for me, using only what I told you.
3. For each tool, judge whether I would notice within a week if it disappeared, and explain why in one line.
4. Identify overlaps: where a tool ranked higher already covers the same job.
5. Rank the tools from most to least earned, then give each a verdict: Keep, Downgrade, Cancel, or Probation (with the job it must prove in 30 days).
Output format: a table with columns Rank, Tool, Job it does for me, Would I notice?, Overlaps with, Verdict. Under the table, list the monthly saving in pounds if I follow every Cancel and Downgrade, showing the sum.
Constraints:
- Base every judgement on my stated usage, not on reputation, reviews or features I did not mention.
- Mark any statement that is an inference rather than something I said with [GUESS].
- Do not quote current prices or plan limits from memory; tell me to check the vendor's pricing page instead.
Before you answer, check: did every verdict follow from my own described use, is every guess labelled, and does the savings sum add up?
This scenario is hypothetical. Sam runs a two-person bookkeeping practice and finds five AI charges on the statement: two general assistants, a meeting transcriber, an image tool and a research search tool.
The table shows one assistant used almost daily for client emails and summarising HMRC guidance; the second opened twice, to "compare answers". The transcriber handled every client call. The image tool made one newsletter header. The research tool was used three times, each for something the main assistant had already answered.
Ranking: main assistant first, transcriber second. Second assistant: Cancel, overlap with slot one. Image tool: Cancel, and use the free tier the one time a quarter it is needed. Research tool: Probation for 30 days with a named job, "source-checked answers for client queries", or it goes too. Sam's spend roughly halves, and the work does not change.
Ranking by excitement: a release or a viral list makes a tool feel essential before it has done a job for you. The close second is keeping overlapping assistants "for comparison", which, unless comparing outputs is your work, is a hobby with a direct debit.
It will not tell you which tool is objectively best or predict next month's pricing. It does not assess data protection; check each vendor's terms separately for client data. And if you describe how you wish you used a tool rather than how you did, the ranking will flatter it.